The following commits could not be applied individually due to context
differences between the monorepo and the public repo's build files.
They have been applied as a cumulative diff to ensure the final state
matches the monorepo exactly:
- chore: sync CI files with 3.0 branch to eliminate merge conflicts (rd-public/tsdb!271)
- revert(refactor): dynamically link taosd taosudf taosmqtt against libtaosnative.so to reduce binary size (revert #183) (rd-public/tsdb!282)
- fix(docs): autofix formatting issues across all doc files (rd-public/tsdb!296)
- feat: support -DBUILD_SANITIZER=true on windows for debug build (rd-public/tsdb!291)
- feat(build): build cache, mirror, and sccache optimizations (rd-public/tsdb!326)
- docs: update image for three replica (rd-public/tsdb!324)
- enh: shared storage on windows (rd-public/tsdb!333)
- fix(cmake): convert ext_libs3 from git clone to URL tarball download (rd-public/tsdb!360)
- feat: dual-source deps and comprehensive docs/packaging (cherry-pick to main) (rd-public/tsdb!352)
- fix(cmake): guard DOWNLOAD_EXTRACT_TIMESTAMP for CMake < 3.24 and fix duplicate Cargo.lock entry (rd-public/tsdb!369)
- fix: test case execution failure in pytest.sh (rd-public/tsdb!338)
- enh: built-in compilation support for Python UDF plugins use abi3 (rd-public/tsdb!325)
* fix: decimal string conversion missing in tRowBuildFromBind2WithBlob
Root cause: tRowBuildFromBind2WithBlob lacked the DECIMAL/DECIMAL64
string-to-binary conversion that exists in tRowBuildFromBind2. When a
table contains both DECIMAL and BLOB columns, the blob code path is
taken (tRowBuildFromBind2WithBlob), which treated DECIMAL as a raw
fixed-size binary type and read 16 bytes directly from the user buffer.
Since the user provides decimal values as text strings (e.g. "21.4300"),
the 15-byte buffer was too small, causing a stack-buffer-overflow.
Fix: Add pSchemaExt parameter to tRowBuildFromBind2WithBlob and add
DECIMAL/DECIMAL64 string-to-binary conversion (decimal128FromStr /
decimal64FromStr) in the fixed-size else branch, mirroring the logic
in tRowBuildFromBind2. Update the call site in parInsertStmt.c to pass
pSchemaExt.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* fix(stmt2): correct DECIMAL in KV+blob row build and align bind path with parsed columns
- tRowBuildKVRowWithBlob / tRowBuildKVRowWithBlob2: copy fixed columns via
VALUE_GET_DATUM() so DECIMAL uses pData instead of the trivial val field.
- tRowBuildFromBind2WithBlob: mirror tRowBuildFromBind2 — accept parsedCols,
correct bufArray indexing with numOfFixedValue, TAOS_CHECK_GOTO/lino, and
free decimal128 heap after each successful row (and on error) to plug leaks.
- parInsertStmt: pass parsedCols into tRowBuildFromBind2WithBlob.
- Add stmt2Case.stmt2_decimal_blob_interleaved in stmt2Test
* fix review
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* fix(scalar): prevent crash when IN expression has invalid type from unhandled node
Root cause: sclGetNodeType() silently set *type = -1 and returned
TSDB_CODE_SUCCESS for unhandled node types. This propagated -1 as
ctx->type.selfType, which was then cast to (uint32_t)-1 = 4294967295
and used as an index into tDataTypes[], causing an out-of-bounds
crash in scalarGenerateSetFromList().
The bug was triggered by semantically odd SQL such as:
ifnull(b not between vb and a, n in (...)) in (today(), today(), now())
where a boolean result is compared against a timestamp list.
Fixes:
1. sclGetNodeType(): return TSDB_CODE_QRY_INVALID_INPUT for unhandled
node types instead of silently setting type = -1.
2. sclInitParam(): validate selfType before using it as list element
type to prevent OOB access in vectorGetConvertType(-1, ...).
3. scalarGenerateSetFromList(): add IS_INVALID_TYPE() guard at entry
as a defensive check against any future invalid-type paths.
Add regression test: test_scalar_invalid_type.py
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* fix: case when and decimal query crash issues
* fix: sclGetNodeType missing cases cause invalid input in query/subquery
Add missing node type cases to sclGetNodeType() that were hit after
ed9de01 changed the fallthrough from silent type=-1 to hard error:
1. QUERY_NODE_LEFT_VALUE (27): return ctx->type.opResType, used when
the planner rewrites constants into ASSIGN(LEFT_VALUE, VALUE)
operators for scalarSup pre-processing (e.g. mavg(2.0, 3))
2. QUERY_NODE_REMOTE_VALUE_LIST (61): read resType from SExprNode,
used when IN subquery RHS is a remote value list node
3. QUERY_NODE_REMOTE_VALUE / QUERY_NODE_REMOTE_ZERO_ROWS /
QUERY_NODE_REMOTE_ROW: read resType from embedded SValueNode,
defensive coverage for other remote node types
Also extend doSetInputDataBlock() dummy-column creation to cover
the first constant argument of indefinite-rows functions (mavg, csum,
diff) so pInput->pData[0] is not NULL at function execution time.
Fixes: select distinct t1,'abc',tbname from st1 -> invalid input
Fixes: 1 in (select 1 from st1) -> invalid input
Fixes: mavg(cast(2 as float), 3) -> invalid input
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Implement week/month/year units for stream PERIOD trigger with natural
boundary alignment and offset support.
Key changes:
- Parser: Add validation for natural time units (w/n/y) and offset parameter
- Time utilities: Add getDuration() support for week/month/year units
- TriggerTask: Implement window calculation with natural boundary alignment
- Week: align to Monday 00:00:00
- Month: align to 1st of month 00:00:00
- Year: align to Jan 1st 00:00:00
- Add offset support: PERIOD(1w, 1d) shifts window by 1 day
- Unit tests: Parser validation, time utilities, TriggerTask window calculation
- System tests: End-to-end tests for week/month/year units with offset
- Documentation: Update user manual with natural time unit examples
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
